The evolution of AI in drug discovery: learning from history’s mistakes (Part 2)

Sujeegar Jeevanandam, an expert with 13 years of experience in life sciences R&D, offers valuable insights into the current state and future trajectory of artificial intelligence in drug discovery.  Having previously drawn parallels with industry adoption of electronic lab notebooks in part 1, this article expands on his vision for the future of the industry, sharing his forward-thinking ideas and practical recommendations.

Catalysts for change

To accelerate the adoption of AI in drug discovery, Sujeegar identifies several key catalysts that could drive meaningful change. He highlights factors that can help bridge the gap between current hesitations and widespread acceptance, ensuring AI becomes a valuable tool for the industry.

These catalysts include:

• Demonstrable evidence: The industry needs clear examples of machine learning models that solve real scientific problems more effectively than traditional methods. These success stories need to come from scientists themselves, highlighting regular, practical applications rather than mere proof-of-concept demonstrations.
• Fear of missing out (FOMO):  As more success stories emerge, the fear of falling behind could drive wider AI adoption, much like it did with the electronic lab notebook (ELN) revolution.
• Strategic implementation: Sujeegar advises organisations, particularly smaller biotech firms, to approach AI adoption patiently and methodically. He recommends focusing on solving specific, high-value problems rather than attempting to transform everything at once. The key is to treat AI as another tool in the problem-solving toolkit rather than a comprehensive mystical solution.

Promising areas for AI implementation

When discussing the most transformative AI applications in drug discovery, Sujeegar identifies a particularly ambitious goal: developing AI models that can simulate human pharmacokinetics and pharmacodynamics (PK/PD) using only preliminary laboratory data. This breakthrough could fundamentally change how drugs are developed and tested. To understand the significance of this goal, it is important to recognise the current drug development process.

Today, researchers must progress through multiple stages of testing: starting with in vitro studies (experiments in test tubes or petri dishes), they then move to animal studies (preclinical testing) before finally advancing to human clinical trials. Each transition between these stages involves significant uncertainty because results from one stage do not always accurately predict outcomes in the next. This uncertainty contributes to the high failure rate and enormous costs of drug development.

Sujeegar envisions AI models that bridge a critical gap by accurately predicting a drug’s behaviour in the human body using early-stage data. Pharmacokinetics describes how the body processes a drug – including its absorption, distribution, metabolism and elimination. In contrast, pharmacodynamics explains what the drug does to the body, including both desired therapeutic effects and potential side effects. Currently, understanding these aspects requires extensive animal testing followed by human trials.

The potential impact of reliable AI-driven PK/PD prediction would be revolutionary in several ways:

First, it would dramatically reduce the need for animal testing, addressing both ethical concerns and the limitations of animal models in predicting human responses. Second, it would allow researchers to identify potential issues far earlier in the development process, before investing in expensive clinical trials. Third, it would enable more efficient design of clinical trials by providing better predictions of effective dosing ranges and potential safety concerns.

Sujeegar points out that clinical trials remain one of the biggest bottlenecks in drug development, both in terms of time and cost. The ability to simulate human responses before beginning animal or toxicology studies could significantly streamline this process. Researchers could focus their resources on candidates with the highest probability of success, design more targeted trials and potentially reduce the number of trial participants needed to demonstrate efficacy and safety.

Nevertheless, developing such complex models presents enormous technical challenges. They must account for the vast complexity of human biology, including variations in genetics, metabolism and environmental factors that can influence drug responses. Additionally, the models must also demonstrate reliable predictive power across different types of drugs and therapeutic areas. Hence, Sujeegar refers to this application as a ‘holy grail’ – a revolutionary breakthrough and   ambitious goal that could transform drug development, yet its realisation requires overcoming significant scientific and technical hurdles.

Recommendations for biotechs

For biotechs beginning their AI journey, Sujeegar emphasises the importance of patience and methodical progress. He recommends:

1. Focusing on data quality and management as much as the research itself
2. Starting with a single, well-defined problem that is worth solving
3. Taking a systematic approach to problem-solving
4. Building trust through successful implementations
5. Avoiding the temptation to ‘boil the ocean’ by trying to implement too much at once.

Conclusion

The integration of AI into drug discovery represents a significant shift in how research is conducted, but success requires careful navigation of technical, cultural and organisational challenges. By learning from historical technological transitions and taking a measured, strategic approach to implementation, organisations can better position themselves to leverage AI’s potential while avoiding common pitfalls. As the industry moves forwards, the key will be finding the right balance between enthusiasm for AI’s possibilities and maintaining a practical mindset regarding the realities of implementation in complex research environments.